Charging pile intelligent detection method and system based on three-dimensional visualization

Abstract

The invention discloses a charging pile intelligent detection method and system based on three-dimensional visualization, and relates to the technical field of data processing. Shooting the target charging pile at multiple angles to obtain a first image set, performing appearance defect detection on the first image set, and marking a defect area to obtain a second image set; performing 3D modeling according to the second image set to obtain an initial three-dimensional visual model, and performing fault positioning on the target charging pile to obtain a fault area; and mapping the fault area to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and outputting an intelligent detection result according to the target three-dimensional visualization model. Through multi-angle shooting, appearance defect detection, 3D modeling and fault positioning, and final target three-dimensional visual model output, fault visual maintenance positioning is realized, the detection accuracy and efficiency are improved, the cost is reduced, and the maintenance speed is accelerated.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a charging pile intelligent detection method and system based on three-dimensional visualization. Background Art

[0002] As an important infrastructure for electric vehicles, charging piles play a key role in providing energy for electric vehicles. With the increase of global environmental awareness and the rapid development of the new energy vehicle industry, the construction and popularization of charging piles have become an important part of promoting green travel and reducing carbon emissions. From the initial simple charging equipment to today's modern charging stations with intelligent detection, efficient charging and remote management functions, charging piles have also entered a stage of rapid development.

[0003] Patent No. CN118358424A discloses a charging data monitoring method for a charging pile. When the charging pile is connected to an electric vehicle, a temperature sensor installed on the charging gun is used to collect the real-time charging temperature of the charging gun at each moment. Based on the real-time charging temperature of the charging gun and the real-time ambient temperature after the charging pile is connected to the electric vehicle, a temperature time series reliability equation is determined. Based on the temperature time series reliability equation, it is determined whether the charging pile has charging abnormalities.

[0004] Although the above technology solves some problems, there are still some problems. For example, the existing charging pile fault detection and positioning is cumbersome, and the fault is not intuitive enough, resulting in slow maintenance and low efficiency. Summary of the invention

[0005] The purpose of the present invention is to solve the problems that the existing charging pile fault detection and positioning is cumbersome and the fault is not intuitive enough, resulting in slow maintenance and low efficiency, and to propose a charging pile intelligent detection method and system based on three-dimensional visualization.

[0006] In a first aspect of the present invention, a charging pile intelligent detection method based on three-dimensional visualization is first proposed, the method comprising: Shooting the target charging pile at multiple angles to obtain a first image set, performing appearance defect detection on the first image set and marking defective areas to obtain a second image set; Performing 3D modeling according to the second image set to obtain an initial three-dimensional visualization model, and locating the fault of the target charging pile to obtain a fault area; The fault area is mapped to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and an intelligent detection result is output according to the target three-dimensional visualization model.

[0007] Optionally, the intelligent detection result includes the fault coordinates of the target charging pile, and locating the fault of the target charging pile to obtain the fault area includes: Acquire the operation signal of the target charging pile, filter the fault signal through a preset filtering algorithm to obtain a target operation signal, and input the target operation signal into a charging pile fault detection model for detection; the charging pile fault detection model includes: a one-dimensional convolution layer, a convolution attention module, a maximum pooling layer, and an average pooling layer; Inputting the target operation signal into a one-dimensional convolution layer to extract fault features to obtain a fault feature data set, and inputting the fault feature data set into a convolution attention module to perform feature optimization selection to obtain a fault feature map; Inputting the fault feature map into the maximum pooling layer and the average pooling layer respectively to aggregate the fault features to obtain a first fault feature map and a second fault feature map, and adding the elements in the first fault feature map and the second fault feature map to obtain a third fault feature map; Determine the channel weight of each element in the third fault feature map to obtain an element weight map, and multiply the element weight map by the element in the fault feature map to obtain a target feature map; The target charging pile is fault-located according to the target characteristic map to obtain the fault coordinates of the fault area.

[0008] Optionally, the intelligent detection result also includes the health status of the target charging pile; after performing 3D modeling according to the second image set to obtain an initial three-dimensional visualization model, and locating the fault of the target charging pile to obtain a fault area, the method further includes: Acquire the operating data of the target charging pile, and extract the features of the operating data through a convolutional neural network model to obtain time series features; the operating data includes: current, voltage and temperature; Obtain historical operation data, input the historical operation data and the time series features into the target BiGRU model for modeling, and obtain a hidden state vector; the hidden state vector contains the entire time series information; The hidden state vector is input into the fully connected layer for prediction to obtain a prediction result, and the health status of the charging pile is determined according to the prediction result.

[0009] Optionally, the historical operation data and the time series features are input into a target BiGRU model for modeling, and the method includes: Step 1: Determine the hyperparameters in the BiGRU model and assign values ​​to the hyperparameters to obtain a hyperparameter search space; Step 2: randomly generate multiple sets of hyperparameter combinations, and use each set of hyperparameter combinations as the initial solution, and evaluate each initial solution to obtain the optimal solution; the evaluation is to calculate the fitness of each solution; Step 3: Perform an update operation to iteratively optimize the hyperparameter combination in each initial solution until the maximum number of iterations is met or the fitness does not change, and output the final solution; Step 4: Determine optimal hyperparameters according to the final solution, and update the BiGRU model according to the optimal hyperparameters to obtain a target BiGRU model.

[0010] Optionally, judging the health status of the charging pile according to the prediction result includes: Calculate a health status assessment coefficient of the charging pile according to the prediction result, and evaluate the target charging pile according to the health status assessment coefficient; If the health status assessment coefficient is greater than the health threshold, or the change in the health status assessment coefficient within a preset time period is greater than the change threshold, it is determined that the target charging pile has a safety hazard and a fault warning is issued to the system.

[0011] In a second aspect of the present invention, a charging pile intelligent detection system based on three-dimensional visualization is proposed, including: an image acquisition module, a three-dimensional modeling module and a fault mapping module: The image acquisition module is used to photograph the target charging pile from multiple angles to obtain a first image set, and to perform appearance defect detection on the first image set and mark the defective area to obtain a second image set; The three-dimensional modeling module is used to perform 3D modeling according to the second image set to obtain an initial three-dimensional visualization model, and to locate the fault of the target charging pile to obtain a fault area; The fault mapping module is used to map the fault area to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and output an intelligent detection result according to the target three-dimensional visualization model.

[0012] Optionally, the three-dimensional modeling module includes: a filtering processing module, a fault feature extraction module, a fault feature aggregation module, a fault feature enhancement module and a fault location output module: The filtering processing module is used to obtain the operation signal of the target charging pile, filter the fault signal through a preset filtering algorithm to obtain a target operation signal, and input the target operation signal into the charging pile fault detection model for detection; the charging pile fault detection model includes: a one-dimensional convolution layer, a convolution attention module, a maximum pooling layer and an average pooling layer; The fault feature extraction module is used to input the target operation signal into a one-dimensional convolution layer to extract fault features to obtain a fault feature data set, and input the fault feature data set into a convolution attention module to perform feature optimization selection to obtain a fault feature map; The fault feature aggregation module is used to input the fault feature map into the maximum pooling layer and the average pooling layer respectively to perform fault feature aggregation to obtain a first fault feature map and a second fault feature map, and add the elements in the first fault feature map and the second fault feature map to obtain a third fault feature map; The fault feature enhancement module is used to determine the channel weight of each element in the third fault feature map to obtain an element weight map, and multiply the element weight map with the element in the fault feature map to obtain a target feature map; The fault location output module is used to locate the fault of the target charging pile according to the target characteristic map to obtain the fault coordinates of the fault area.

[0013] Optionally, the system further comprises: an operation data feature extraction module, a hidden state vector acquisition module and a prediction result output module: The operating data feature extraction module is used to obtain the operating data of the target charging pile, and extract the features of the operating data through a convolutional neural network model to obtain time series features; the operating data includes: current, voltage and temperature; The hidden state vector acquisition module is used to acquire historical operation data, input the historical operation data and the time series features into the target BiGRU model for modeling, and obtain a hidden state vector; the hidden state vector contains the entire time series information; The prediction result output module is used to input the hidden state vector into the fully connected layer for prediction to obtain a prediction result, and judge the health status of the charging pile according to the prediction result.

[0014] Optionally, the hidden state vector acquisition module includes: a hyperparameter assignment module, a hyperparameter evaluation module, a parameter updating module and a model updating module: The hyperparameter assignment module is used to determine the hyperparameters in the BiGRU model and assign values ​​to the hyperparameters to obtain a hyperparameter search space; The hyperparameter evaluation module is used to randomly generate multiple sets of hyperparameter combinations, and use each set of hyperparameter combinations as an initial solution, and evaluate each initial solution to obtain an optimal solution; the evaluation is to calculate the fitness of each solution; The parameter updating module is used to perform an updating operation, iteratively optimize the hyperparameter combination in each initial solution until the maximum number of iterations is met or the fitness does not change, and output a final solution; The model updating module is used to determine the optimal hyperparameters according to the final solution, and update the BiGRU model according to the optimal hyperparameters to obtain a target BiGRU model.

[0015] Optionally, the prediction result output module includes: an evaluation coefficient calculation module and a health status evaluation module: The evaluation coefficient calculation module is used to calculate the health status evaluation coefficient of the charging pile according to the prediction result, and evaluate the target charging pile according to the health status evaluation coefficient; The health status assessment module is used to determine that the target charging pile has a safety hazard and issue a fault warning to the system if the health status assessment coefficient is greater than the health threshold, or the change in the health status assessment coefficient within a preset time period is greater than the change threshold.

[0016] Beneficial effects of the present invention: The present invention proposes a charging pile intelligent detection method based on three-dimensional visualization, which obtains a first image set by shooting a target charging pile from multiple angles, performs appearance defect detection on the first image set and marks the defect area to obtain a second image set; performs 3D modeling based on the second image set to obtain an initial three-dimensional visualization model, and performs fault location on the target charging pile to obtain the fault area; maps the fault area to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and outputs intelligent detection results based on the target three-dimensional visualization model. By shooting from multiple angles, appearance defect detection, 3D modeling and fault location, and then outputting the final target three-dimensional visualization model, visual maintenance and location of faults are achieved, the accuracy and efficiency of detection are improved, the cost is reduced, and the maintenance speed is accelerated. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the accompanying drawings.

[0018] Figure 1 A flowchart of a charging pile intelligent detection method based on three-dimensional visualization is provided for an embodiment of the present invention; Figure 2 A framework diagram of a charging pile intelligent detection system based on three-dimensional visualization is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the present invention, the description of "first", "second", etc. is only used for descriptive purposes, and cannot be understood as indicating or implying its relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0021] The embodiment of the present invention provides a charging pile intelligent detection method based on three-dimensional visualization. Figure 1 , Figure 1 A flowchart of a charging pile intelligent detection method based on three-dimensional visualization provided by an embodiment of the present invention. The method comprises the following steps: S101, photographing a target charging pile from multiple angles to obtain a first image set, performing appearance defect detection on the first image set and marking defective areas to obtain a second image set; S102, performing 3D modeling according to the second image set to obtain an initial 3D visualization model, and locating the fault of the target charging pile to obtain a fault area; S103, mapping the fault area to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and outputting an intelligent detection result according to the target three-dimensional visualization model.

[0022] An intelligent detection method for charging piles based on three-dimensional visualization provided by an embodiment of the present invention realizes visual repair and positioning of faults through multi-angle shooting, appearance defect detection, 3D modeling and fault location, and finally outputs a target three-dimensional visualization model, thereby improving the accuracy and efficiency of detection, reducing costs and speeding up repairs.

[0023] In one implementation, by photographing the target charging pile from multiple angles and performing appearance defect detection and marking on it, the appearance information of the charging pile can be fully captured, and the existing defective areas can be accurately identified and marked; that is, multiple cameras are deployed around the target charging pile, and the multiple cameras capture images of the target charging pile at the same time (or video streams, images extracted from the video stream at the same time); appearance defect detection is a process of analyzing and processing the charging pile images in the first image set through image processing technology and machine learning algorithms. Automatic identification of appearance defects of the charging pile can be achieved through image preprocessing, feature extraction, classifier training and other steps. In the above process, the algorithm will learn the characteristics of the normal appearance of the charging pile and compare them with the actual features in the image, so as to identify the existing defective areas and mark them.

[0024] In one implementation, 3D modeling is performed based on the marked image set to generate an initial three-dimensional visualization model of the charging pile. This step allows the structure and shape of the charging pile to be intuitively displayed in three-dimensional space. 3D modeling is a process of three-dimensionally reconstructing the charging pile by utilizing the defective area information marked in the second image set, combined with computer vision and graphics technology. The three-dimensional point cloud data of the charging pile can be generated by applying feature extraction, matching and three-dimensional reconstruction algorithms to the images, and then through meshing, texture mapping and other steps, the initial three-dimensional visualization model of the charging pile is finally obtained; wherein, the three-dimensional visualization model includes the area with damaged appearance.

[0025] In one implementation, the fault area is mapped to the initial three-dimensional visualization model to obtain the target three-dimensional visualization model, and the model is output as the intelligent detection result, which helps maintenance personnel to quickly understand the fault condition of the charging pile, improve maintenance efficiency, and optimize the management process; from multi-angle shooting, appearance defect detection to 3D modeling and fault location, and then to the target three-dimensional visualization model output, a comprehensive, accurate and intuitive detection of the appearance defects of the charging pile is achieved, the accuracy and efficiency of the detection are improved, the cost and difficulty of manual detection are reduced, and the management efficiency is optimized for the operation and maintenance management of the charging pile.

[0026] In one embodiment, the intelligent detection result includes the fault coordinates of the target charging pile, and the fault location of the target charging pile to obtain the fault area includes: In one embodiment, the intelligent detection result includes the fault coordinates of the target charging pile, and the fault location of the target charging pile to obtain the fault area includes: Obtain the operation signal of the target charging pile, filter the fault signal through a preset filtering algorithm to obtain the target operation signal, and input the target operation signal into the charging pile fault detection model for detection; the charging pile fault detection model includes: a one-dimensional convolution layer, a convolution attention module, a maximum pooling layer and an average pooling layer; The target operation signal is input into the one-dimensional convolution layer to extract fault features to obtain a fault feature dataset, and the fault feature dataset is input into the convolution attention module to perform feature optimization selection to obtain a fault feature map; The fault feature map is respectively input into the maximum pooling layer and the average pooling layer to aggregate the fault features to obtain a first fault feature map and a second fault feature map, and the elements in the first fault feature map and the second fault feature map are added to obtain a third fault feature map; Determine the channel weight of each element in the third fault feature map to obtain an element weight map, and multiply the element weight map with the element in the fault feature map to obtain a target feature map; The fault coordinates of the fault area are obtained by locating the fault of the target charging pile according to the target feature map.

[0027] In one implementation, the preset filtering algorithm is a minimum fourth-order moment adaptive filtering algorithm, which is a signal processing method that adaptively adjusts filter parameters by minimizing the fourth-order moment cost function of the error signal. The algorithm can use the filter parameter results of the previous moment to automatically adjust the filter parameters at the current moment to adapt to the unknown or time-varying statistical characteristics of the signal and noise, thereby achieving optimal filtering.

[0028] In one implementation, the non-Gaussian and nonlinear nature of charging pile faults will reduce the accuracy of fault detection. The distortion of current leads to the introduction of high-order harmonics. It is necessary to extract effective fault feature vectors through signal processing technology in order to clearly distinguish various faults, thereby obtaining operating signal information of charging pile fault characteristics that are not significant. The minimum fourth-order moment adaptive filtering algorithm can significantly reduce noise interference and improve the accuracy of fault detection. It is suitable for charging pile fault detection and can clearly distinguish complex fault characteristics such as high-order harmonics caused by input current distortion.

[0029] In one implementation, the basic model of the charging pile fault detection model is a convolutional neural network model; the one-dimensional convolutional neural network bidirectional gated loop unit detection model is combined with the minimum fourth moment adaptive filtering algorithm to achieve accurate and fast charging pile fault state detection, and a support vector machine is used as the output layer of the model to ensure the detection effect of the charging pile fault state. That is, the charging pile operation signal after one-dimensional filtering is input into the improved one-dimensional convolution layer to extract fault features, the fault feature signal is input into the bidirectional gated loop unit to further learn the time series, the fault feature is input into the support vector machine for feature classification, and the fault state detection result of the charging pile is output.

[0030] In one implementation, the processed target operation signal is input into the charging pile fault detection model. The convolutional neural network model includes: a one-dimensional convolution layer, a convolutional attention module, a maximum pooling layer, and an average pooling layer. The one-dimensional convolution layer is responsible for extracting fault features, continuously extracting higher-dimensional feature information from the filtered charging pile operation signal, and generating a fault feature data set; the convolutional attention module usually performs two types of pooling operations on the input feature map (here, the feature map of the filtered charging pile operation signal) in the channel attention module: maximum pooling and average pooling. These two pooling operations extract different spatial information from the input feature map respectively. After these two pooling operations, two new feature maps are obtained, which respectively represent the aggregated representation of the input feature map under different spatial information; the two feature maps obtained by the pooling operation are added element by element, the purpose is to fuse the spatial information extracted by the two pooling operations, so as to obtain a third fault feature map that integrates the two types of information; the channel weights of each element in the third fault feature map are determined to obtain an element weight map, that is, the third fault feature map is processed through a shared multi-layer perceptron (MLP). MLP usually includes a hidden layer and an output layer to further enhance the dimension of features and extract key channel weights (each element in the feature map represents the importance weight of the corresponding channel). The element weight map (channel weight) is multiplied with the elements in the fault feature map to weight each channel in the input fault feature map according to the channel weight, thereby enhancing the channel information of fault detection (fault information of charging piles).

[0031] In one implementation, the fault feature map is input into the maximum pooling layer and the average pooling layer respectively to aggregate the fault features. The pooling layer includes the maximum pooling layer and the average pooling layer. After the convolution, it is input into the pooling layer. By processing and counting the fault characteristics of the charging pile in a certain area, feature selection and dimensionality reduction are achieved, which can effectively reduce the number of network parameters and avoid network overfitting.

[0032] In one implementation, the above charging pile fault detection process, from the acquisition of operating signals to filtering processing, and then to the extraction, optimization, aggregation and location of fault features, forms an efficient and accurate fault detection system. The system can not only quickly identify the fault area of ​​the charging pile, but also accurately analyze and locate the fault features.

[0033] In one embodiment, the intelligent detection result also includes the health status of the target charging pile; after performing 3D modeling according to the second image set to obtain an initial three-dimensional visualization model, and locating the fault of the target charging pile to obtain the fault area, the method further includes: Obtain the operating data of the target charging pile, and extract the features of the operating data through the convolutional neural network model to obtain time series features; the operating data includes: current, voltage and temperature; Obtain historical operation data, input the historical operation data and time series features into the target BiGRU model for modeling, and obtain a hidden state vector; the hidden state vector contains the entire time series information; The hidden state vector is input into the fully connected layer for prediction to obtain the prediction result, and the health status of the charging pile is judged according to the prediction result.

[0034] In one implementation, the convolutional neural network model (CNN) is used to extract features from the operating data of the charging pile, which can automatically learn and capture key information in the data, such as time series features such as current, voltage, and temperature. These features reflect the various physical states of the charging pile during operation and are an important basis for subsequent prediction of the health status. The efficient feature extraction capability of CNN enables the model to understand the operating status of the charging pile more accurately. Time series features: refers to the characteristics shown in data arranged in chronological order, such as periodicity, trend, etc.

[0035] In one implementation, the target BiGRU model is used to model the historical operation data and the extracted time series features, which can simultaneously consider the forward and backward dependencies in the charging pile operation data. The BiGRU model can capture the information in the entire time series, including historical status and future trends, through its unique bidirectional structure, thereby generating a hidden state vector containing the information of the entire time series; the historical operation data and time series features are input into the target BiGRU model for modeling to obtain the hidden state vector. For example: there is now a set of charging pile operation data, each time point contains dimensional data such as current, voltage and temperature, and the time points are connected in a string to form a time series; when there are 100 time points in the time series, the features at each time point are input into the BiGRU. The BiGRU will process these features step by step and update its hidden state at each time point. When the 100th time point (the last step of the time series) is processed, the hidden state of the BiGRU will contain the complete information of the entire time series.

[0036] In one implementation, the hidden state vector is input into the fully connected layer for prediction, and the health status prediction result of the charging pile can be obtained. The fully connected layer can output a prediction result (failure rate and other operating data) representing the health status of the charging pile by further processing the hidden state vector. Based on this prediction result, it can be determined whether the charging pile is in a normal state, whether there is a potential fault or maintenance is required: Fully connected layer: A layer structure in a neural network, where each neuron is connected to all neurons in the previous layer, and is used to further process and classify the input data.

[0037] In one embodiment, historical operation data and time series features are input into a target BiGRU model for modeling, and the method includes: Step 1: Determine the hyperparameters in the BiGRU model and assign values ​​to the hyperparameters to obtain a hyperparameter search space; Step 2: randomly generate multiple sets of hyperparameter combinations, and use each set of hyperparameter combinations as the initial solution, and evaluate each initial solution to obtain the optimal solution; the evaluation is to calculate the fitness of each solution; Step 3: Perform an update operation to iteratively optimize the hyperparameter combination in each initial solution until the maximum number of iterations is met or the fitness does not change, and output the final solution; Step 4: Determine the optimal hyperparameters based on the final solution, and update the BiGRU model based on the optimal hyperparameters to obtain the target BiGRU model.

[0038] In one implementation, when the fault of the charging pile is determined, a health prediction analysis of the charging pile is performed based on the operating data of the charging pile; the performance of the BiGRU model (bidirectional gated recurrent unit, which can simultaneously capture the forward and backward dependencies in the sequence data by combining the gated recurrent unit layers in the forward and backward directions, thereby enhancing the model's understanding and processing capabilities of the sequence data) depends on the setting of its hyperparameters. The hyperparameters include the convolution kernel size, the number of hidden units of the GRU, the learning rate, etc. The selection of these parameters directly affects the training effect and prediction accuracy of the model; the FHO algorithm, as an optimization algorithm based on swarm intelligence, has a strong global search capability and fast convergence speed, and can automatically search for the optimal hyperparameter combination, thereby improving the performance of the BiGRU model; reducing the cost of manual parameter adjustment and enhancing the generalization ability of the model.

[0039] In one implementation, a hyperparameter search space is constructed by specifying the hyperparameters in the BiGRU model and assigning values ​​to them (establishing a range or a set of possible values ​​for each hyperparameter), ensuring that the search process can cover all possible parameter combinations, so that there is a chance to find the global optimal solution; multiple sets of hyperparameter combinations are randomly generated as initial solutions, and the fitness of each set of solutions (such as accuracy, loss function value, etc.) is calculated, that is, its performance on a specific task is evaluated. This process helps to quickly screen out parameter combinations with better performance.

[0040] In one implementation, the hyperparameter combination in the initial solution is iteratively optimized and the parameter values ​​are continuously adjusted to find a better solution. When the maximum number of iterations is reached or the fitness no longer changes significantly, the final solution is output. This process ensures the stability and effectiveness of the hyperparameter optimization process; based on the optimal hyperparameters in the final solution, the BiGRU model is updated to obtain the target BiGRU model. This step optimizes the model performance, so that the target model performs better on specific tasks, and improves the accuracy and generalization ability of the model.

[0041] In one implementation, an update operation is performed, for example, by generating falcons and prey to continuously update the hyperparameter combination. The distance between the falcons and the prey is calculated, and the falcon's territory is determined. The falcons update their hyperparameter combinations based on their positions, and the prey inside and outside the territory also update their combinations. By evaluating the fitness value of the updated hyperparameter combination and updating the global optimum to the current best solution, the FHO (Fire Hawk) algorithm gradually searches for improved hyperparameter combinations, thereby improving the performance of the BiGRU model through iterative optimization.

[0042] In one embodiment, judging the health status of the charging pile according to the prediction result includes: The health status assessment coefficient of the charging pile is calculated based on the prediction results, and the target charging pile is evaluated based on the health status assessment coefficient; If the health status assessment coefficient is greater than the health threshold, or the change in the health status assessment coefficient within a preset time period is greater than the change threshold, it is determined that the target charging pile has a safety hazard and a fault warning is issued to the system.

[0043] In one implementation, the health status assessment coefficient of the charging pile is calculated based on the prediction results, that is, the operation data of the charging pile in the future, the failure rate, electrical parameters (voltage, current), the operating temperature of the charging pile, and the operating time of the charging pile are used to fit the health status assessment coefficient; among them, the failure rate: the offline and failure records of the charging pile during the operation and maintenance cycle can reflect the stability and reliability of the charging pile; the operating time of the charging pile: the cumulative operating time of the charging pile from the time it is put into use to the current time, reflecting the service life and aging degree of the charging pile; electrical parameters (voltage, current, etc.): current: the current value output by the charging pile during the charging process, reflecting the load condition of the charging pile and the battery charging status, voltage: the output voltage of the charging pile, which is an important indicator for evaluating whether the charging pile is working normally; the operating temperature of the charging pile: the temperature of each component of the charging pile: including the temperature of the key components such as the internal circuit board of the charging pile, the temperature of the charging gun, and the charging module. Excessive temperature may mean that the charging pile is at risk of overheating.

[0044] In one implementation, the calculation formula of the health status assessment coefficient is: , where F is the health status assessment coefficient (score), V is the voltage, is the standard voltage (at the lower limit of the normal working range), I is the current, is the standard current (at the lower limit of the normal working range), G is the failure rate per unit time (failure rate refers to the number of failures of the charging pile within a certain period of time, for example: the charging pile has 3 failures in a month, and the total operating time of the charging pile in that month is 300 hours, then the failure rate of the charging pile can be calculated as 3 times / 300 hours = 0.01 times / hour), is the maximum acceptable value of the failure rate per unit time (for example: 10 times / hour), T is the operating temperature, is the optimal operating temperature of the charging pile, and t is the operating time of the charging pile; the above parameters have been dimensionless. Send a fault warning to the system, for example: a fault alarm reminds the operation and maintenance personnel to perform maintenance.

[0045] Based on the same inventive concept, the embodiment of the present invention also provides a charging pile intelligent detection system based on three-dimensional visualization. Figure 2 , Figure 2 A schematic diagram of the structure of a charging pile intelligent detection system based on three-dimensional visualization provided by an embodiment of the present invention includes: an image acquisition module, a three-dimensional modeling module and a fault mapping module: An image acquisition module, used for photographing the target charging pile at multiple angles to obtain a first image set, performing appearance defect detection on the first image set and marking the defective area to obtain a second image set; A three-dimensional modeling module, used to perform 3D modeling according to the second image set to obtain an initial three-dimensional visualization model, and to locate the fault of the target charging pile to obtain a fault area; The fault mapping module is used to map the fault area to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and output the intelligent detection result according to the target three-dimensional visualization model.

[0046] An intelligent detection system for charging piles based on three-dimensional visualization provided by an embodiment of the present invention realizes visual repair and positioning of faults through multi-angle shooting, appearance defect detection, 3D modeling and fault location, and finally outputs a target three-dimensional visualization model, thereby improving the accuracy and efficiency of detection, reducing costs and speeding up repairs.

[0047] In one embodiment, the three-dimensional modeling module includes: a filtering processing module, a fault feature extraction module, a fault feature aggregation module, a fault feature enhancement module and a fault location output module: A filtering processing module is used to obtain the operation signal of the target charging pile, filter the fault signal through a preset filtering algorithm to obtain the target operation signal, and input the target operation signal into the charging pile fault detection model for detection; the charging pile fault detection model includes: a one-dimensional convolution layer, a convolution attention module, a maximum pooling layer and an average pooling layer; A fault feature extraction module is used to input the target operation signal into a one-dimensional convolution layer to extract fault features to obtain a fault feature data set, and input the fault feature data set into a convolution attention module to perform feature optimization selection to obtain a fault feature map; A fault feature aggregation module is used to input the fault feature map into the maximum pooling layer and the average pooling layer respectively to perform fault feature aggregation to obtain a first fault feature map and a second fault feature map, and add the elements in the first fault feature map and the second fault feature map to obtain a third fault feature map; A fault feature enhancement module is used to determine the channel weight of each element in the third fault feature map to obtain an element weight map, and multiply the element weight map with the element in the fault feature map to obtain a target feature map; The fault location output module is used to locate the fault of the target charging pile according to the target feature map to obtain the fault coordinates of the fault area.

[0048] In one embodiment, the system further includes: an operation data feature extraction module, a hidden state vector acquisition module and a prediction result output module: The operating data feature extraction module is used to obtain the operating data of the target charging pile and extract the features of the operating data through the convolutional neural network model to obtain the time series features; the operating data includes: current, voltage and temperature; The hidden state vector acquisition module is used to obtain historical operation data, input the historical operation data and time series features into the target BiGRU model for modeling, and obtain the hidden state vector; the hidden state vector contains the entire time series information; The prediction result output module is used to input the hidden state vector into the fully connected layer for prediction to obtain the prediction result, and judge the health status of the charging pile according to the prediction result.

[0049] In one embodiment, the hidden state vector acquisition module includes: a hyperparameter assignment module, a hyperparameter evaluation module, a parameter update module and a model update module: The hyperparameter assignment module is used to determine the hyperparameters in the BiGRU model and assign values ​​to the hyperparameters to obtain the hyperparameter search space; The hyperparameter evaluation module is used to randomly generate multiple sets of hyperparameter combinations, and use each set of hyperparameter combinations as the initial solution, and evaluate each initial solution to obtain the optimal solution; the evaluation is to calculate the fitness of each solution; The parameter update module is used to perform the update operation, iteratively optimize the hyperparameter combination in each initial solution until the maximum number of iterations is met or the fitness does not change, and output the final solution; The model updating module is used to determine the optimal hyperparameters according to the final solution, and update the BiGRU model according to the optimal hyperparameters to obtain the target BiGRU model.

[0050] In one embodiment, the prediction result output module includes: an evaluation coefficient calculation module and a health status evaluation module: An evaluation coefficient calculation module is used to calculate the health status evaluation coefficient of the charging pile according to the prediction result, and evaluate the target charging pile according to the health status evaluation coefficient; The health status assessment module is used to determine that the target charging pile has safety hazards and issue a fault warning to the system if the health status assessment coefficient is greater than the health threshold, or the change in the health status assessment coefficient within a preset time period is greater than the change threshold.

[0051] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A charging pile intelligent detection method based on three-dimensional visualization, characterized in that: The method comprises: Shooting the target charging pile at multiple angles to obtain a first image set, performing appearance defect detection on the first image set and marking defective areas to obtain a second image set; Performing 3D modeling according to the second image set to obtain an initial three-dimensional visualization model, and locating the fault of the target charging pile to obtain a fault area; The fault area is mapped to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and an intelligent detection result is output according to the target three-dimensional visualization model.

2. The intelligent detection method of charging pile based on three-dimensional visualization according to claim 1 is characterized in that: The intelligent detection result includes the fault coordinates of the target charging pile, and the fault location of the target charging pile to obtain the fault area includes: Acquire the operation signal of the target charging pile, filter the fault signal through a preset filtering algorithm to obtain a target operation signal, and input the target operation signal into a charging pile fault detection model for detection; the charging pile fault detection model includes: a one-dimensional convolution layer, a convolution attention module, a maximum pooling layer, and an average pooling layer; Inputting the target operation signal into a one-dimensional convolution layer to extract fault features to obtain a fault feature data set, and inputting the fault feature data set into a convolution attention module to perform feature optimization selection to obtain a fault feature map; Inputting the fault feature map into the maximum pooling layer and the average pooling layer respectively to aggregate the fault features to obtain a first fault feature map and a second fault feature map, and adding the elements in the first fault feature map and the second fault feature map to obtain a third fault feature map; Determine the channel weight of each element in the third fault feature map to obtain an element weight map, and multiply the element weight map by the element in the fault feature map to obtain a target feature map; The target charging pile is fault-located according to the target characteristic map to obtain the fault coordinates of the fault area.

3. The intelligent detection method of charging pile based on three-dimensional visualization according to claim 2 is characterized in that: The intelligent detection result also includes the health status of the target charging pile; After performing 3D modeling according to the second image set to obtain an initial three-dimensional visualization model and locating the fault of the target charging pile to obtain a fault area, the method further includes: Acquire the operating data of the target charging pile, and extract the features of the operating data through a convolutional neural network model to obtain time series features; the operating data includes: current, voltage and temperature; Obtain historical operation data, input the historical operation data and the time series features into the target BiGRU model for modeling, and obtain a hidden state vector; the hidden state vector contains the entire time series information; The hidden state vector is input into the fully connected layer for prediction to obtain a prediction result, and the health status of the charging pile is determined according to the prediction result.

4. The intelligent detection method of charging pile based on three-dimensional visualization according to claim 3 is characterized in that: Inputting the historical operation data and the time series features into a target BiGRU model for modeling, the method comprising: Step 1: Determine the hyperparameters in the BiGRU model and assign values ​​to the hyperparameters to obtain a hyperparameter search space; Step 2: randomly generate multiple sets of hyperparameter combinations, and use each set of hyperparameter combinations as the initial solution, and evaluate each initial solution to obtain the optimal solution; the evaluation is to calculate the fitness of each solution; Step 3: Perform an update operation to iteratively optimize the hyperparameter combination in each initial solution until the maximum number of iterations is met or the fitness does not change, and output the final solution; Step 4: Determine optimal hyperparameters according to the final solution, and update the BiGRU model according to the optimal hyperparameters to obtain a target BiGRU model.

5. The intelligent detection method of charging pile based on three-dimensional visualization according to claim 3 is characterized in that: Judging the health status of the charging pile according to the prediction result includes: Calculate a health status assessment coefficient of the charging pile according to the prediction result, and evaluate the target charging pile according to the health status assessment coefficient; If the health status assessment coefficient is greater than the health threshold, or the change in the health status assessment coefficient within a preset time period is greater than the change threshold, it is determined that the target charging pile has a safety hazard and a fault warning is issued to the system.

6. A charging pile intelligent detection system based on three-dimensional visualization, characterized in that: The system includes: an image acquisition module, a three-dimensional modeling module and a fault mapping module: The image acquisition module is used to photograph the target charging pile from multiple angles to obtain a first image set, and to perform appearance defect detection on the first image set and mark the defective area to obtain a second image set; The three-dimensional modeling module is used to perform 3D modeling according to the second image set to obtain an initial three-dimensional visualization model, and to locate the fault of the target charging pile to obtain a fault area; The fault mapping module is used to map the fault area to the initial three-dimensional visualization model to obtain a target three-dimensional visualization model, and output an intelligent detection result according to the target three-dimensional visualization model.

7. The charging pile intelligent detection system based on three-dimensional visualization according to claim 6 is characterized in that: The three-dimensional modeling module includes: a filtering processing module, a fault feature extraction module, a fault feature aggregation module, a fault feature enhancement module and a fault location output module: The filtering processing module is used to obtain the operation signal of the target charging pile, filter the fault signal through a preset filtering algorithm to obtain a target operation signal, and input the target operation signal into the charging pile fault detection model for detection; the charging pile fault detection model includes: a one-dimensional convolution layer, a convolution attention module, a maximum pooling layer and an average pooling layer; The fault feature extraction module is used to input the target operation signal into a one-dimensional convolution layer to extract fault features to obtain a fault feature data set, and input the fault feature data set into a convolution attention module to perform feature optimization selection to obtain a fault feature map; The fault feature aggregation module is used to input the fault feature map into the maximum pooling layer and the average pooling layer respectively to perform fault feature aggregation to obtain a first fault feature map and a second fault feature map, and add the elements in the first fault feature map and the second fault feature map to obtain a third fault feature map; The fault feature enhancement module is used to determine the channel weight of each element in the third fault feature map to obtain an element weight map, and multiply the element weight map with the element in the fault feature map to obtain a target feature map; The fault location output module is used to locate the fault of the target charging pile according to the target characteristic map to obtain the fault coordinates of the fault area.

8. The charging pile intelligent detection system based on three-dimensional visualization according to claim 7 is characterized in that: The system also includes: an operation data feature extraction module, a hidden state vector acquisition module and a prediction result output module: The operating data feature extraction module is used to obtain the operating data of the target charging pile, and extract the features of the operating data through a convolutional neural network model to obtain time series features; the operating data includes: current, voltage and temperature; The hidden state vector acquisition module is used to acquire historical operation data, input the historical operation data and the time series features into the target BiGRU model for modeling, and obtain a hidden state vector; the hidden state vector contains the entire time series information; The prediction result output module is used to input the hidden state vector into the fully connected layer for prediction to obtain a prediction result, and judge the health status of the charging pile according to the prediction result.

9. The charging pile intelligent detection system based on three-dimensional visualization according to claim 8 is characterized in that: The hidden state vector acquisition module includes: a hyperparameter assignment module, a hyperparameter evaluation module, a parameter update module and a model update module: The hyperparameter assignment module is used to determine the hyperparameters in the BiGRU model and assign values ​​to the hyperparameters to obtain a hyperparameter search space; The hyperparameter evaluation module is used to randomly generate multiple sets of hyperparameter combinations, and use each set of hyperparameter combinations as an initial solution, and evaluate each initial solution to obtain an optimal solution; the evaluation is to calculate the fitness of each solution; The parameter updating module is used to perform an updating operation, iteratively optimize the hyperparameter combination in each initial solution until the maximum number of iterations is met or the fitness does not change, and output a final solution; The model updating module is used to determine the optimal hyperparameters according to the final solution, and update the BiGRU model according to the optimal hyperparameters to obtain a target BiGRU model.

10. The charging pile intelligent detection system based on three-dimensional visualization according to claim 8, characterized in that: The prediction result output module includes: an evaluation coefficient calculation module and a health status evaluation module: The evaluation coefficient calculation module is used to calculate the health status evaluation coefficient of the charging pile according to the prediction result, and evaluate the target charging pile according to the health status evaluation coefficient; The health status assessment module is used to determine that the target charging pile has a safety hazard and issue a fault warning to the system if the health status assessment coefficient is greater than the health threshold, or the change in the health status assessment coefficient within a preset time period is greater than the change threshold.

Images